Diffusion-Reward Adversarial Imitation Learning